**5. Results**

**5.2. The role of variables**

Even though the graphical view did provide evidence for the convergence and
persistence, it didn’t provide any quantitative evidence. In this section the importance
of the variables of *Rajan & Zingales (1995)*, *Frank & Goyal (2007)*, and *Lemmon, Roberts *

*& Zender (2008)*described earlier will be further examined. To estimate the effects the

following equation is applied20_{: }

*Leverage*

*it*

*=**α + β X*

*t-1*

*+γ Leverage*

*i0*

*+ V*

*t*

*+ ε*

*it*

The abbreviations *i* & *t *represent firms and years. *X* represents a set of 1-year lagged
control variables. *Leveragei0* is firm *i*’s initial leverage defined as the first non-missing

**43 **

**43**

**Figure 7: Leverage patterns in event time according to unexpected leverage portfolios. **The sample
consists of all non-financial firm-year observations in the Compustat database & DataStream database
from 1989 to 2010. Each panel consists of four lines representing the average unexpected leverage of
four portfolios including low, medium, high & very high across event time. The process is as follow:
first, for the independent variables log of assets, profitability, growth ratio and tangibility 1-year
lagged variables are created followed by estimating cross-sectional regression for each calendar year
whereas industries are also included. Per year from the outcomes of cross-sectional regression the
actual leverage are tracked leading to a certain amount that is unexplained, which is termed
“unexpected Leverage”. In other words unexpected leverage is residuals from cross-sectional
regression on the variable named earlier. Year 1990 is the portfolio formation period. In 1990 firms
are ranked according to their unexpected leverage and are divided into four portfolios. As these
portfolio compositions is kept constant across the event time, the average leverage is calculated for
the subsequent 11 years (1991-2001). The process of ranking and averaging is repeated for the firms
all the way through 1999, leading to a total of 10 sets across 11 event times. Finally for every event
time the average of portfolios is calculated leading to four final portfolios per event time. Panel A and
B represent unexpected book leverage portfolios for all firms and the survivors. Panel C and D
represent unexpected market leverage portfolios for all firms and the survivors.

value for leverage given a firm since existence21_{. }_{Y}_{ measures the importance of firms’ }

21_{ First encountered observation that is found. }

-.2 -.1 0 .1 .2 BO O K LEVER AG E 0 1 2 3 4 5 6 7 8 9 10 11 EVENT TIME (Years)

Low Medium High Very High Panel A: Unexpected Book Leverage Portfolios

-.2 -.1 0 .1 .2 MAR KET L EVER AG E 0 1 2 3 4 5 6 7 8 9 10 11 EVENT TIME (Years)

Low Medium High Very High Panel C: Unexpected Market Leverage Portfolios

-.2 -.1 0 .1 .2 BO O K LEVER AG E 0 1 2 3 4 5 6 7 8 9 10 11 EVENT TIME (Years)

Low Medium High Very High Panel B: Unexpected Book Leverage Portfolios(Surv.)

-.2 -.1 0 .1 .2 MAR KET L EVER AG E 0 1 2 3 4 5 6 7 8 9 10 11 EVENT TIME (Years)

Low Medium High Very High Panel D: Unexpected Market Leverage Portfolios(Surv.)

**44 **

**44**

initial leverage.* V *is a year fixed effect and ϵ is a random error that is heteroskedastic
and correlated within firms. The model is executed for book leverage and market
leverage. Also the same method is applied to the Survivors. As the OLS regression
starts only with the variable Initial leverage, it is continuously extended. First the
determinants log of assets, growth ratio, profitability, and tangibility are added
followed by a second expansion where the variables median industry leverage, cash
flow volatility and dividend dummy variable are added*.* Further to make comparison
possible each coefficient is scaled by corresponding variable’s standard deviation.
The results represent the estimated effect of one standard deviation (*σ*) change in
vector *X* on leverage. Table 3 represents the findings. In Panel A all firms are taken
into account and in Panel B only firms with at least 11 years of existence.

The second and the fourth columns in Panel A represent the results for the OLS
regression when only initial leverage is considered. A one standard deviation change
in initial book leverage corresponds to an average change of 8% in future value of
book leverage. For the market model this effect is even 9%, and is significant for both
models at the 1% level. The total variability explained by this variable amounts 23%
(11%) in book (market) leverage model. In the original study the result for this
coefficient is 7% (11%) for book (market) leverage and show an adjusted R-squared of
13% (20%). This suggests that although the magnitude in variability explained differ
between this study and the original study, the initial leverage of firms contain a
certain permanent component which is in line with earlier graphical findings. When
the model is extended with four traditional determinants of *Rajan & Zingales (1995) *of
which the results are presented in columns 3 and 6 some features are worth noting.
First as the effect of initial leverage in the book leverage model remains equally
strong it losses some of its strong feature in the market model. Except for the growth
ratio which is not significant all other variables are equally important in making
predictions for market leverage and are in term of sign highly significant. Finally
when the determinants of *Frank & Goyal (2007) *are added to the model, initial
leverage losses some of its magnitude in book leverage model and is supplement with
growth ratio which has an equal effect. In this book leverage model specification
median industry leverage is the variable with the highest marginal effect.
Considering the same specification leads to a dramatic loss in the coefficient
magnitude of initial market leverage and its statistical and economic importance.
Also in the market model the most single important determinant is now the median
industry leverage with an effect of 9% change in market leverage when one standard
deviation change occurs in median industry leverage.

**45 **

**45**

In this study the log of assets is positively related to leverage in both models. These results being albeit not always significant are in line with thoughts behind the trade- off and signaling theory implying that as firm gets bigger in size diversification may

**Table 3: Contribution of initial leverage for forecasting purpose. ** The sample consists of all non-
financial firm-year observations in the Compustat database & DataStream database from 1989 to
2010. Each column represents scaled parameter estimates from OLS regressions on book and market
leverage on the underlying key specified determinants including initial book leverage, initial market
leverage, log of assets, growth ratio, profitability, tangibility, industry median leverage, cash flow
volatility, dividend, and year fixed effects. When a column contains yes for the dummy variable year
fixed effects, it implies that calendar year fixed effects are included in the regression. The italic and
bold values represent the t-statistics which are robust to both clustering at the firm level and
heteroscedasticity. Regarding interpretation, a one standard deviation change in underlying key
specified determinant is associated with the presented value for parameter estimate change in
leverage. Panel A represent the findings of all firms whereas panel B takes the survivor sample into
account. Further each column includes the amount of observation and the adjusted R-squared. For the
construction of the variables consult 4.2. Variable construction. Italic and bold values of 2.58 or higher,
representing t-statistics, are highly significant at the 1% level for a two-tailed test. Values above 1.96
are significant at the 5% level.

Panel A: All Firms

Variable Book Leverage Market Leverage

Initial leverage 0.08 0.07 0.04 0.09 0.05 0.01
**8.49 ****7.62 ****4.59 ****6.20 ****4.09 *** 0.85 *
Log Of Assets 0.02 0.01 0.04 0.03

**2.57**

**1.75**

**3.12***Growth Ratio 0.04 0.04 0.01 0.01*

**2.43**

**5.58**

**5.29**

**1.00***Profitability -0.01 -0.01 -0.05 -0.04*

**0.14**

**-1.80**

**-1.61**

**-4.18***Tangibility 0.04 0.03 0.05 0.02*

**-3.61**

**4.69**

**3.39**

**3.89**

**1.72**Median Industry Leverage 0.07 0.09

**8.78 ****8.86 **

Cash Flow Volatility 0.00 -0.01

**0.76 ****-0.70 **

Dividend Payer -0.01 -0.02

**-2.11 ****-2.52 **

Year Fixed Effects No Yes Yes No Yes Yes

Adj.*R2* _{0.23 } _{0.29 } _{0.40 } _{0.11 } _{0.19 } _{0.30 }

**46 **

**46**

(Continued)

Panel B: Survivors

Variable Book Leverage Market Leverage

Initial leverage 0.08 0.06 0.03 0.07 0.04 0.01
**6.62 ****6.18 ****3.62 ****4.46 ****3.22 *** 0.78 *
Log Of Assets 0.02 0.01 0.04 0.03

**2.39**

**1.60**

**3.32***Growth Ratio 0.04 0.04 0.01 0.01*

**2.10**

**5.52**

**5.25**

**1.15***Profitability -0.02 -0.01 -0.06 -0.04*

**0.69**

**-1.70**

**-1.31**

**-3.93***Tangibility 0.05 0.03 0.05 0.02*

**-3.39**

**4.62**

**3.03**

**3.57**

**1.42**Median Industry Leverage 0.06 0.09

**7.03 ****7.89 **

Cash Flow Volatility 0.00 -0.01

**0.06 ****-0.65 **

Dividend Payer -0.00 -0.02

**-2.04 ****-1.96 **

Year Fixed Effects No Yes Yes No Yes Yes

Adj.*R2* 0.21 0.31 0.41 0.09 0.20 0.31

Obs. 2153 1995 1610 1982 1837 1485

take place which lowers the risk of bankruptcy and make borrowing easy. Further,
more debt in combination with large firms are considered as a sign of vitality to the
market *(Paulo et al. 2007)*. On average the marginal effect found for this variable
shows that irrespective of the type of financial system as firms increase in their size
the information asymmetry declines and a positive relationship arises. Even if not
significant in both models growth ratio is positively related to leverage. According to
the pecking order theory growth opportunities have a positive impact on debt when
they are greater than the profits retained, and a negative influence when they are less
than retained profits. On average the growth ratio and the profits retained ratio in
this study amount 0.10 and -0.06 supporting this theory. The positive relationship
found in this study and the negative relationship found in the original study is
assumed to have been caused due to this difference. Another possible explanation lies
in different way of connectedness among the financial systems. In bank oriented
financial systems banks provide most of credit to the economy and in order to
minimize risk they like to hold their contacts as close as possible resulting in high
connectedness and low information asymmetry. Dutch companies operating in bank

**47 **

**47**

oriented financial system are more likely to have this close lenders and borrowers
relationships. On the other hand in market oriented finical systems companies raise
funds in capital markets. It is assumed that the degree of connectedness is much
lower (i.e. information asymmetry is high) and that managers are possibly forced to
reduce their leverage ratios because according to the trade-off theory the cost of
financial distress rises with expected growth. Profitability is only found to be
significant in the market model. Also these results are in line with pecking order
theory who assumes that the most profitable companies turn to self-financing rather
than debt financing. The same (i.e. similar) finding also hold for companies being
active in market-based financial systems. The positive coefficient of the variable
tangibility implies that companies’ fixed assets serve as collateral security for outside
capital *(Myer & Majluf, 1984)*. Even if expected the marginal effect to be more
prominent in bank oriented financial systems the results provide the opposite.
Remarkably also here on average one finds the effect of this variable to be nearly
equal among both finical systems when focusing on the market model.

The effect of the median industry leverage is as expected and found by *Frank & Goyal *

*(2007)*, and *Lemmon, Roberts & Zender (2008)*. In this study cash flow volatilities are

found to be not a relevant factor in explaining the capital structure as these are not
statistically significant. Earlier for dividend one found that on average 54% to 59% of
the Dutch non-financial firms pay dividend to their shareholders. From pecking order
theory perspective firms become less levered over time as the amount of dividend
paid to the shareholders is reduced* Frank & Goyal (2007)*. A comparison between the
average dividend paid before and after 1999 supports this statement. Before 1999 on
average 71% of Dutch non-financial firms paid dividend. This percentage amounts
42% after 1999. The marginal effect found for this variable is more prominent in the
original study (i.e.-3% (-5%) for book (market) model). This implies that the value that
dividend signals is much higher in original study. It is in general known that Dutch
firms have a high ownership concentration which is assumed less likely to be the case
in US. Earlier one argued that in bank oriented financial systems closely tied
relationships are more likely. Based on these arguments one can argue that the value
of the signal being low in this study is possibly due to this difference (i.e. information
asymmetry is low). For the Survivors in Panel B the results are nearly similar.

In sum as important the variable initial leverage in the book leverage model it is not equally important in making projections for market leverage. But one should consider when interpreting the results that initial leverage is a constant value that does not change over time (time-invariant). Regarding other traditional determinants change in time implies variation in these variables and so these are considering the time

**48 **

**48**

effect as well. It is rather interesting to see how the coefficients of initial leverage behave as one extends the model with other determinants and calendar year observations. The graphical view in figure 8 represents the magnitude of the coefficient of initial leverage across event time. The black lines represent the market model and the red lines take the book model into account. Each line represents the estimated standardized coefficients from an OLS regression of book and market leverage on several different specifications, starting with initial leverage. The panel data is continuously extended with calendar year observations. The first prediction is based on data from 1989-1994. The same method is applied for the rest of the sample leading to a total of seventeen estimated and standardized coefficients per line.

**Figure 8: Magnitude of coefficient across event time according to different specification. **The figure
represents six lines. Each line maps the magnitude of the coefficient of initial leverage on future
leverage over time an according to different specifications. The INML line takes only the effect of initial
market leverage into account and it estimates the standardized coefficient from an OLS regression of
market leverage. Lines equipped with abbreviations R&Z and F&G take also the variables of Rajan &
Zingales and Frank & Goyal into account. The first estimation is made for observations until 1994.
Every time when this method is applied one year is added to the sample (starting with 1994…2010). All
coefficients are found to be significant (except for specification 3 of market leverage model) and
robust to both clustering at the firm level and heteroscedasticity. The result shows how important and
volatile initial leverage is when different specification and time period (calendar year observations) are
considered. INBL line represents the effect on book leverage.

0 .05 .1 .15 .2 .25 M agn itud e O f C oe ffic ien t 1995 2000 2005 2010 Event Time(Years)

INML INML+R&Z INML+R&Z+F&G

INBL INBL+R&Z INBL+R&Z+F&G

**49 **

**49**

The graphical view provides interesting features. When the model only considers
initial leverage, the results in this study show that initial leverage is the most
important determinant of future capital structure in the market model. As this model
is expanded with the traditional determinants of *Rajan & Zingales (1995)* the
coefficient in the market model undergoes a loss of 53.33% in magnitude. For the
book leverage this amount a loss of 18%. Both lines outweigh the estimated effect of
the same coefficient as the model is expanded with the traditional determinants,
implying that traditional determinants in overall affect this variable negatively.
Although this is normal in multiple regression analysis as the independent variables
effect one another, a bigger surprise is the difference in effect of initial leverage on
market leverage and book leverage when the event time is lengthened (i.e. calendar
year observations are added). The coefficient of initial leverage in the market model
and book model declines on average by 70% and 18% implying that initial leverage is
more stable determinant of book leverage with an average within standard deviation
of 0.01 in contrast to market model which has an average within standard deviation
of 0.03. Interestingly when one continuously extend the third specification model for
market leverage where all traditional determinates are considered with calendar year
observations 1998 and further on, one find a loss in statistical sign for the coefficient
of the variable Initial market leverage. In all other cases this variable is found to be
highly significant.

The findings indicate that in order to solve the capital structure puzzle one should also look at time invariant factors. For certain time variant factors one not only find a different sign but also a different measure of strength, implying that there are certain differences among type of economies. Initial leverage being only a fraction of a bunch of time invariant firm specific factors indicate in this study that on average this variable contains equally or sometimes even more information about corporate leverage compared to the time variant factors. Further these results provide evidence that even time invariant factors have different effect over time and within different specifications. These results however do not indicate the economic importance of variables as these are combined. In other words existing determinants as a whole may have a different impact on capital structure in contrast to their individual impact. In the next paragraph these issues will be elaborated.